GeneFlow: Translation of Single-cell Gene Expression to Histopathological Images via Rectified Flow

2025-10-31

AI Summary

This study addresses the challenging cross-modal generation task of synthesizing high-resolution histopathological images (e.g., H&E, DAPI) from single-cell gene expression data to uncover associations between transcriptional states and cellular morphology/spatial interactions. To this end, we propose GeneFlow—a novel framework built upon rectified flow that establishes a continuous bijective mapping, effectively circumventing the inherent many-to-one bottleneck in gene-to-image translation. GeneFlow integrates an attention-based RNA encoder, a conditional U-Net backbone, and a high-order ODE solver to enable high-fidelity, perturbable (genetic or chemical) image synthesis. Compared to diffusion-based baselines, GeneFlow achieves significant improvements across multiple quantitative metrics. Generated images faithfully recapitulate subcellular structures—including nuclear-cytoplasmic features—and preserve spatial neighborhood relationships. Moreover, GeneFlow successfully identifies disease-specific imaging phenotypes, demonstrating its utility for interpretable, biology-guided digital pathology.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Spatial transcriptomics (ST) technologies can be used to align transcriptomes with histopathological morphology, presenting exciting new opportunities for biomolecular discovery. Using ST data, we construct a novel framework, GeneFlow, to map transcriptomics onto paired cellular images. By combining an attention-based RNA encoder with a conditional UNet guided by rectified flow, we generate high-resolution images with different staining methods (e.g. H&E, DAPI) to highlight various cellular/tissue structures. Rectified flow with high-order ODE solvers creates a continuous, bijective mapping between transcriptomics and image manifolds, addressing the many-to-one relationship inherent in this problem. Our method enables the generation of realistic cellular morphology features and spatially resolved intercellular interactions from observational gene expression profiles, provides potential to incorporate genetic/chemical perturbations, and enables disease diagnosis by revealing dysregulated patterns in imaging phenotypes. Our rectified flow-based method outperforms diffusion-based baseline method in all experiments. Code can be found at https://github.com/wangmengbo/GeneFlow.

Problem

• Mapping transcriptomics to histopathological images via rectified flow
• Generating realistic cellular morphology from gene expression profiles
• Addressing many-to-one relationship between transcriptomics and image manifolds

Innovation

• Attention-based RNA encoder maps transcriptomics to images
• Rectified flow creates bijective mapping between gene and image manifolds
• Conditional UNet generates high-resolution histopathological images from gene expression